The semiarid US Department of Energy Hanford site has a deep vadose zone with low recharge rates. Contaminants originating from nuclear waste processing are expected to move slowly through the vadose zone. The movement of certain contaminants can be facilitated by colloids. We hypothesized that the low recharge rates and low water contents in semiarid regions, however, tend to inhibit movement of colloidal particles, thereby reducing the risk for colloid-facilitated contaminant transport. The goal of this study was to investigate whether in situ natural colloids can be mobilized and transported in undisturbed, deep vadose zone sediments at the Hanford site under typical, semiarid recharge rates. We sampled an undisturbed sediment core (i.d. 50 cm, 59.5 cm height) from a depth of 17 m below ground at the Hanford 200 Area. The core was set up as a laboratory lysimeter and exposed to an infiltration rate of 18 mm/yr by applying simulated pore water onto the surface. Particle concentrations were quantified in the column outflow, and selected samples were examined microscopically and for elemental composition (transmission electron microscopy and energy dispersive X-ray). Measured water contents and potentials were used to calibrate a numerical model (HYDRUS-1D), which was then applied to simulate colloid mobilization from the sediment core. During 5.3 years of monitoring, natural colloids like silicates, aluminosilicates, and Fe-oxides were observed in the core outflow, indicating the continuous mobilization of in situ colloids. The total amount of particles mobilized during 5.3 years corresponded to 1.1% of the total dispersible colloids inside the core. Comparison of the amounts of colloids released with weathering rates suggests that mineral weathering can be a major source of the mobilized colloids. The fitted colloid release rate coefficient was 6 to 7 orders of magnitude smaller than coefficients reported from previous studies, where disturbed Hanford sediments and higher flow rates were used. Our findings demonstrate that even under low recharge rates and water contents typical for semiarid, deep vadose zone sediments, particles can continuously be mobilized, although the total mass of particles is low.
 The US Department of Energy (US DOE) Hanford site served as a Pu processing facility from 1943 to 1989. During the decades of Pu production, large amounts of radioactive nuclear waste (including Cs, Am, and U) were generated. Considerable amounts of the nuclear waste were directly discharged to the ground, while much of the highly radioactive waste was stored in massive tanks buried below ground surface [Gephart, 2003].
 Leakage from the waste tanks has been confirmed first in 1959 [Zachara et al., 2007]. Radioactive isotopes of elements like Cs, Sr, Tc, Pu, and Am were detected in the tank wastes from three of Hanford's largest tanks SX-108, BX-102, and T-106 [Jones et al., 2000a, 2000b, 2001]. Leaked tank wastes reacted with the surrounding sediments, causing dissolution and precipitation of minerals [Mashal et al., 2004, 2005a, 2005b; Deng et al., 2006]. The more mobile tank contaminants, such as anionic , and U(VI), have been reported to migrate to groundwater at the Hanford site [Knepp, 2002a, 2002b; Myers, 2005]. Less mobile contaminants, such as cationic radionuclides, can be potentially transported via colloidal particles [Zhuang et al., 2003; Chen et al., 2005; Cheng and Saiers, 2010].
 Many radionuclides are not very mobile due to adsorption reactions on sediment surfaces or precipitation in relatively insoluble solid phases. However, when the associated solid phase falls in the colloidal size range, particle transport can be an important mechanism of subsurface contaminant transport [McCarthy and Zachara, 1989; Honeyman, 1999]. Colloid and colloid-facilitated radionuclide transport has been intensively investigated at the Hanford site [Flury et al., 2002; Cherrey et al., 2003; Zhuang et al., 2003, 2004; Chen and Flury, 2005; Chen et al., 2005; Czigany et al., 2005a]. These studies showed that both native and neo-formed colloids can facilitate the migration of radionuclides such as 137Cs, and that colloid-facilitated transport can occur under both saturated and unsaturated flow conditions. Decreasing water content has been shown to reduce colloid transport as colloids are trapped in water films. However, transient flow conditions can lead to more colloid mobilization caused by moving air-water interfaces, which can scour colloids from surfaces [Saiers and Lenhart, 2003; Zhuang et al., 2007; Shang et al., 2008; Aramrak et al., 2011]. Particularly, infiltration into a relatively dry soil or sediment causes pronounced colloid mobilization [Cheng and Saiers, 2009, 2010]. Also, sequences of intermittent irrigation pulses can enhance the amount of mobilized colloids [Zhuang et al., 2007].
 The Hanford site is characterized by a Mediterranean, semiarid climate [Harvey, 2000]. In this climate, while winter precipitation causes substantial water infiltration near the soil surface, groundwater recharge rates are low [Hsieh et al., 1973; Gee, 1987; Gee et al., 1992, 1994, 2005a]. Using different methods to estimate recharge rates (lysimeters, chloride mass balance), Gee and coworkers have reported recharge rates ranging from near zero to 100 mm/yr [Gee et al., 1992, 2005b]. Model simulations performed for the Hanford tank farms showed that as recharge rate decreased (50, 30, and 10 mm/yr), peak arrival time of contaminants was delayed and the peak concentration was reduced [Khaleel et al., 2007].
 Colloid and colloid-facilitated contaminant transport has been documented from laboratory experiments using disturbed Hanford sediments [Cherrey et al., 2003; Chen et al., 2005; Cheng and Saiers, 2010]. A lysimeter field study at the Hanford site showed that in situ colloids and Eu could be transported from the soil surface down to 2.1 m depth within 2.5 months under natural and forced rainfall conditions [Liu et al., 2013]. While colloid and Eu movement in this field study were likely caused by transients in the near-surface water flow [Liu et al., 2013], under deep vadose zone conditions, with steady-state and low recharge rates, colloid transport is likely less pronounced. No information, however, is available about colloid movement in the deep vadose zone at Hanford.
 Based on previous laboratory results obtained with disturbed Hanford sediments, we know that colloid transport and mobilization is strongly impacted by water content or flow rate—the lower the water content or flow rate, the less colloids leach from columns [Cherrey et al., 2003; Chen et al., 2005; Shang et al., 2008]. Similar results were found in other porous materials [Kaplan et al., 1993; Saiers and Lenhart, 2003; Gao et al., 2006]. Cumulative amounts of colloids mobilized from packed Hanford sediments were about 50 times smaller at a water content of 0.21 cm3/cm3 than at 0.32 cm3/cm3 [Shang et al., 2008]. Colloid breakthrough curves at a water content of 0.11 cm3/cm3 not only showed five times smaller peak concentration than at 0.4 cm3/cm3, but also considerable tailing [Cherrey et al., 2003]. These results suggest that colloid transport and mobilization under the low water contents in the deep vadose zone at Hanford will be limited, but nonetheless possible.
 The objectives of this study were to (1) test whether natural colloidal materials can be mobilized and translocated under conditions typical for the deep vadose zone at the semiarid Hanford site, and (2) to quantify the extent of colloid mobilization. We hypothesized that under the low and steady-state recharge rate at the semiarid Hanford site, colloid mobilization and transport is limited, but nonetheless existent. We monitored water flow and colloid transport using an undisturbed sediment core, collected from 17 m below ground at the Hanford site, and mimicked the natural recharge under controlled laboratory conditions.
2. Materials and Methods
2.1. Undisturbed Sediment Core
 An undisturbed sediment core was collected on 8 March 2003 from the Hanford Environmental Restoration Disposal Facility (ERDF), which is located between the Hanford 200 West and East areas (Figure 1). The core was taken from an uncontaminated layer of the sand-dominated facies association of Hanford Formation sediments from a depth of 17 m below ground surface (Figure 2a). A flat, horizontal bench of sediment was prepared by digging into the slanted wall of the ERDF pit. Then an intact, undisturbed core sample was taken by using a stainless steel cylinder (10 GA T-304, i.d. 50 cm, height 59.5 cm). A front loader with a 3 m wide bucket was used to push the cylinder into the sediments (Figure 2b). When the core was completely inserted into the sediments, the sediments around the core were excavated, and a beveled stainless steel plate was pushed along the bottom of the core to shave the core off the underlaying sediments (Figure 2c). About 17 L of liquid nitrogen were poured onto the surface of the core to freeze the top of the core to provide stability for transportation. A wooden plate was tightened to the top of the cylinder, and the core was moved to Washington State University (Figure 2d).
 The core originates from the sand-dominated facies association of the Pleistocene Hanford Formation sediments [Pace et al., 2003; Reidel and Chamness, 2007]. The sediments at the sampling location had fine- and coarse-textured layers, and our core was sampled out of a coarse-textured layer (Figure 2e). These coarse-textured sands consist mainly of illite, smectite, kaolinite, vermiculite, mica, quartz, feldspars, and pyroxene [Mashal et al., 2004]. The colloidal size minerals include smectite, kaolinite, illite, and quartz [Czigany et al., 2005b]. The bulk density of the sediments was 1.6 g/cm3 and the porosity was 0.396 cm3/cm3 [Shang et al., 2008]. The particle size distribution was measured with the hydrometer method, including both physical and chemical pretreatment steps [Gee and Bauder, 1986], and the results are shown in Table 1.
Table 1. Particle Size Distribution for Five Different Depth Increments
 The amount of total dispersible colloids was determined by leaching a packed sediment column (i.d. 4.3 cm, length 22.5 cm) with eight pore volumes of irrigation solution (see section 2.3) under sprinkling conditions. The amount of dispersible colloids was measured gravimetrically and expressed as mass of colloids dispersed per mass of sediments (Table 1). This amount of dispersible colloids is much less than the amount of the clay fraction of the sediments, because the dispersion procedure of the particle size analysis is much more vigorous.
2.2. Experimental Setup
 The sediment core in the stainless steel column was set up as a lysimeter (Figure 2f) in a dark coldroom at 12.4 ± 0.3°C, corresponding to the average air temperature at the Hanford site [Hanford Meteorological Station, 2012]. We assumed that the temperature at 17 m depth, where the core was taken, equals the average air-temperature. A schematic of the experimental setup is shown in Figure 3. Five porous cup tensiometers (i.d. 0.6 mm, length 2.9 cm, Soil Moisture Equipment Corp.) and five, custom-made 3-rod Time Domain Reflectometry (TDR) probes (diameter of the rods = 2.5 mm, length of the rods = 20 cm) were inserted into the column to monitor matric potential and volumetric water content, respectively. The tensiometers and TDR probes were installed oppositely at depths 9.5, 19.5, 29.5, 39.5, and 49.5 cm. The tensiometers were fitted with pressure transducers (PX26, Omega Engineering, Stamford, CT) and connected to a data logger (CR10X, Campbell Scientific Inc., Logan, UT). The TDR probes were connected to a TDR-100 and a CR10X data logger (Campbell Scientific Inc.). Water potentials and water content were monitored hourly. Tensiometers were calibrated with a hanging water column, and periodic recalibrations were made to correct for drift. The TDR probes were calibrated using packed sediments with controlled water contents. Specific calibrations curves were developed for each tensiometer and TDR probe.
 A Plexiglass plate, containing 12 sections fitted with porous membranes, was attached to the bottom of the column. Each section had a drain connected to a Tygon tube, through which a hanging water column could be applied. The membranes had a bubbling pressure of 90 cm-H2O, corresponding to a pore size of 32 μm, which is big enough to allow colloidal particles to pass through. However, the membranes did not provide the suction as expected, and functioned only as a seepage face open to the atmosphere. We therefore later on installed a fiberglass wick (12.5 mm diameter, Catalog no. 1381, Peperell, MA) 3 cm above the bottom on 30 June 2010 (day 1800). The fiberglass wick was mounted on a cut-open polyvinyl chloride tube (i.d. 3/4 in.) and inserted radially through the entire diameter of the column. A 57 cm hanging piece of the wick provided tension for collecting outflow (Figure 3), and the wick worked as intended. Outflow was collected in glass vials (from the 12 sections of the bottom plate), and in a 250 mL polyethylene bottle (from the wick).
 The entire column was placed on a load cell (AL H22-1K, capacity 500 kg, resolution 100 g, Indiana Scale Co.) which was monitored daily. Two ultraviolet (UV) lamps (Spectroline X-Series, 254 nm, Spectronics Corp, New York) were positioned close to the column surface and turned on three times a day for 60 min to prevent microbial growth at the surface.
2.3. Irrigation and Infiltration
 The column was irrigated uniformly by using a peristaltic pump (Ismatec CP 78001-20, Glattbrugg, Switzerland) and a sprinkler. The sprinkler consisted of 48 Teflon tube drippers (Western Analytical, i.d. 0.15 mm) arranged in a uniform pattern. Irrigation started in July 2005 with a sprinkling rate of 100 mL/d, corresponding to an irrigation rate of 0.5 mm/d (= 182 mm/yr). As this rate was too low for continuous irrigation, sprinkling was intermittent (every 35 min we sprinkled for a period of 5 min). The irrigation water consisted of 0.5 mM NaBr, 0.3 mM KCl, and 0.4 mM CaCl2, adjusted to pH 8 with NaOH, to mimic the chemical composition of Hanford pore water [Serne et al., 2002a].
 The irrigation rate was chosen to fall within the range of annual rainfall at the Hanford site [Hanford Meteorological Station, 2012]. Based on the mass balance of the sediment core, we determined the actual infiltration rate to be 0.05 mm/d (= 18.25 mm/yr), which corresponds to irrigation minus evaporation rates. This infiltration rate represents a low recharge rate typical for the Hanford site; recharge rates have reported to range from 0 to 100 mm/yr [Gee, 1987; Gee et al., 2005b]. First outflow from the column was observed in March 2008. Outflow occurred from all of the outflow sections initially, but later on, several sections ceased to yield outflow. Due to this inconsistent outflow pattern, we decided to install a fiber glass wick in June 2010, which, after an initial spike of outflow, then collected consistent outflow.
2.4. Outflow Characterization and Data Analysis
 Column outflow water was analyzed for electrophoretic mobility (ZetaSizer 3000HSa, Malvern Instruments Ltd., Malvern, UK), pH, electrical conductivity, and particle concentration with UV/visible light spectrometer (VIS) absorbance at 280 nm (Ocean Optics USB4000-UV-VIS, Ocean Optics Inc., Dunedin, FL). Absorbance was translated to particle concentrations by using a calibration curve developed from a colloid stock suspension. The detection limit for the absorbance measurements was 1.83 mg/L, determined following the procedure described in Skoog et al. .
 The stock suspension was obtained by leaching a packed sediment column (i.d. 4.3 cm, length 22.5 cm) with eight pore volumes of irrigation solution to dislodge mobile colloids. The mass of mobilized colloids was determined gravimetrically. The stock suspension was then diluted to obtain calibration standards and to develop a calibration curve of colloid mass concentration versus UV-Vis absorbance. The stock suspension contained natural particles that varied in size and composition, which is representative of the colloids collected from the sediment core.
 Selected outflow samples from the sediment core were analyzed by transmission electron microscopy (TEM) and energy dispersive X-ray analysis (EDX) (JEOL 1200 EX Transmission Electron Microscope; Philips CM-200). For these microscopic measurements, the outflow samples were sonicated, and a drop of solution was placed on a carbon-nickel microscopy stub and air dried.
 The hourly monitoring data for water potential and water content were averaged on a daily basis, and the time series data then smoothed with a 25 point, second-order Savitzky-Golay algorithm. This procedure eliminated noise from the data, but preserved the general shape of the curves.
2.5. Water Flow Modeling
 We used HYDRUS-1D (Version 4.14, Simunek et al. ) to analyze the water flow and colloid transport in our experiments. HYDRUS was only used to simulate the part of the experiments before the wick was installed. The unsaturated hydraulic properties were parameterized by the van Genuchten-Mualem equations [van Genuchten, 1980]. The modeling domain was discretized in 0.5 cm spaced nodes, with observation points selected at the position where the sensors were installed. An additional observation point was located at the bottom of the core. The initial condition was given by the measured matric potentials at the observation points, with linear interpolation between the points. The modeling was done in a two-stage fashion: we first optimized the hydraulic parameters using the measured water contents and matric potentials (from day 1 to 1400), and we then used the model with the optimized hydraulic parameters to simulate colloid movement.
 The upper boundary condition was selected as constant flow with the measured infiltration rate. In a previous study, we also modeled the water flow with a time-dependent upper boundary condition, following the intermittent sprinkling, and we found no differences compared with constant flow [Vogs, 2009]. The lower boundary was set as a seepage face with a 2 cm-H2O head. Auxiliary tests showed that the bottom plate with the membranes had a resistance corresponding to about 2 cm-H2O head.
 The model was run in inverse mode to determine the unsaturated hydraulic properties of the sediments. For the van Genuchten-Mualem model, we fitted the parameters α, n, and Ks, with m = 1−1/n, for each of the observation nodes. The core was divided in five layers, each having distinct hydraulic properties in terms of the fitting parameters. Residual and saturated water contents (θr and θs) were fixed according to measured values, as was the parameter l = 0.5. Initial values for the parameter estimation were chosen from Vogs . Parameters were then determined by inverse modeling using the Levenberg-Marquardt algorithm with data from the beginning of the irrigation until day 1400, the time period for which we had continuous data for matric potential and water content. Water content and water potential data were weighted by their respective standard deviations as implemented in HYDRUS-1D.
2.6. Colloid Transport Modeling
 For the colloid transport modeling, we assumed the the dispersible colloids to be distributed in five layers according to our measurements (Table 1), and we assumed that the colloids are initially attached to the stationary sediments. We further assumed that, as water flows through the core, colloids can be mobilized by a first-order kinetic colloid release [Shang et al., 2008]:
where C represents the colloid concentration suspended in the aqueous phase (mg/cm3), S is the colloid concentration attached to the sediments (mg/g), t is time (day), ρ is the bulk density (g/cm3), z denotes the coordinate parallel to the flow direction (cm), and β is the first-order colloid release rate coefficient (1/d). A zero flux condition was used at the upper boundary and a zero gradient was used at the bottom boundary . The amount of colloids present in the column for mobilization was and , where S0 was determined by the method described in section 2.1 and numerical values are listed in Table 1. We considered that reattachment of mobilized colloids is negligible.
 The transport model (equations (1) and (2)) was coupled with the water flow model described above within HYDRUS-1D. Simulated cumulative colloid outflow from the model was fitted to the experimental colloid outflow data by adjusting the first-order colloid release rate coefficient β.
3. Results and Discussion
3.1. Water Potentials, Water Contents, and Outflow
 The water monitoring data are summarized in Figure 4. The initial condition of the sediments was close to a no-flow equilibrium, with the water contents ranging from 0.08 to 0.1 cm3/cm3. After 200 days of irrigation, the rate of increase in water potentials and water contents dropped considerably. The values of the top two water content sensors remained fairly constant, but the three lower sensors indicate that the core then wetted up from the bottom. This was because the porous membranes in the bottom plate did not yield outflow, indicating that their intended purpose of draining water under suction was not working properly, likely because of air trapping under the membranes causing a break in the capillary connection. After the first outflow was recorded, the top two sensors (water potential and content) indicate drying of the core. After the fiber glass wicks were installed at day 1800, the core started to drain at a higher rate, mainly draining out the lower part of the core, as indicated by the TDR sensor at 49.5 cm depth.
 Overall, the water potential and water content data show that core was not always under steady-state conditions during the duration of the experiment. After the initial infiltration period, the water potentials varied between −10 and −15 cm-H2O and the water contents varied between 0.08 and 0.15 cm3/cm3 (corresponding to gravimetric water contents of 0.05 to 0.09 g/g) over a time span of 3.5 years. Although we were not able to maintain constant water potentials and water contents throughout the entire experimental period, steady-state periods were obtained for time periods of several months up to 1 year at a time (Figure 4). Water contents measured from sediment cores at the Hanford 200 area are in the range of 0.03 to 0.1 g/g, with occasional values up to 0.15 g/g [Serne et al., 2002b]; therefore, the water content variations in our core were within the range of field-measured water contents.
 The simulated matric potentials followed the experimental data well, while the water contents showed more deviations between model and data, particularly for the lower two sensors (Figure 5). The overall regression coefficient of the model fit was R2 = 0.987. The initial infiltration phase was not fitted as well as the later phase, quasi steady-state conditions. The estimated model parameters are listed in Table 2.
Table 2. Van Genuchten-Mualem Parameters Used for Unsaturated Water Flow Simulationa
The lower and upper limits of the optimized parameters indicate the 95% confidence interval.
θr: measured; θs = 0.394 cm3/cm3 for all layers, measured; l = 0.5 for all layers.
3.2. Colloid Transport
3.2.1. General Observations
 Chemical and colloidal characteristics of the outflow are shown in Figure 6. The pH of the outflow remained fairly constant between pH 7.5 and 8.5, but showed a lower value after the wick was installed, followed by an increase to pH 8.5. Electrolytic conductivity varied between 500 and 2000 μS/cm, corresponding to ionic strengths of 7.2 to 29 mmol/L estimated with the Marion-Babcock equation [Sposito, 2008]. Electrolytic conductivity values are in the range of those reported from pore waters from Hanford sediment cores from the 200 Area; Serne et al. [2002a] lists values of 200 to 6700 μS/cm for pore water from Hanford Formation sediments. The critical coagulation concentration for in situ colloids from Hanford sediments has been reported to be 1.7 to 3.8 mmol/L for calcium dominated pore waters [Czigany et al., 2005b]. As the ionic strength in our core exceed the critical coagulation concentration, we do not expect colloid dispersion to be pronounced. Because of variability in size and composition, some colloids, as indicated by lower electrophoretic mobilities, may form stable suspensions. These are the colloids that we believe are leached from the sediment core.
 The first outflow samples had particle counts below 10 kCounts/s, which is within the background noise of the dynamic light scattering and, no electrophoretic mobility measurements could be made for such low particle counts. In 2009, the particle counts increased up to 50 kCounts/s and electrophoretic mobility measurements were made for these samples (Figure 6c), but have to be considered with caution, as the particle counts still did not meet the required minimum value for accurate measurements (>50 kCounts/s). Electrophoretic mobility values, nonetheless, yielded reasonable results for Hanford colloids (−1 to −3 (μm/s)/(V/cm)) [Shang et al., 2008].
3.2.2. Colloid Concentrations and Flow Rates
 The particle concentrations in the outflow were generally 50 mg/L, with occasionally higher concentrations up to 400 mg/L, and we observed a continuous release of colloids after the first outflow was collected (Figure 6d). The particle concentrations in the outflow were higher than we had expected based on previous reports. In a previous study using the same coarse Hanford sediments as used in this study, Shang et al.  investigated colloid release from packed sediments under different flow rates varying from 0.018 to 0.288 cm/min (= 259 to 4147 mm/d), and observed a positive relationship between the amount of colloids released and flow rate. Colloid concentration in column outflow for a flow rate of 518 mm/d were reported to be 5 to 150 mg/L [Shang et al., 2008]. The lowest flow rates of Shang's study were 3 to 4 orders of magnitude larger than the flow rate in our study here. Based on the strong dependency of the amount of colloid released on flow rate (or water content) observed by Shang et al. , we would have expected much smaller colloid concentrations in our outflow samples.
 Using coarse Hanford sediments similar to ours, Cherrey et al.  investigated the effect of different water saturations (and flow rates) on colloid transport. Cherrey et al.  did not study in situ colloid mobilization, but rather colloid transport through disturbed sediments where colloidal suspensions were passed through packed sediment column under different water contents and flow rates. They showed that colloids were more and more retained inside the sediments as the volumetric water content decreased from saturation to 0.11 cm3/cm3 (corresponding to flow rates of 59,000 to 72 mm/d, respectively). Transport was strongly reduced, but still occurred at a water content of 0.11 cm3/cm3 [Cherrey et al., 2003], a water content similar to the one in our undisturbed sediment core, although the flow rates in Cherrey et al.  were 3 orders of magnitude higher.
 Similar results of decreasing colloid transport and mobilization with decreasing water content or flow rates in Hanford sediments have been reported by others [Chen et al., 2005; Cheng and Saiers, 2010]. Gamerdinger and Kaplan  studied colloid transport and deposition in unsaturated sand and Yucca mountain tuff, and they observed that decreasing water content resulted in more colloid deposition, thus less colloid transport.
 Using a field lysimeter, Kaplan et al.  applied a one time irrigation of 51 mm/h for 2 h onto a lysimeter surface and monitored water outflow with a zero-tension drainage port at 1.5 m depth. They observed colloid concentrations in the outflow ranging from 300 to 1700 mg/L. In a field study with an agricultural soil, Villholth et al.  monitored particle concentrations in drain lines installed at 1.1 m depth. They applied drip irrigation 12 mm/h for 3 h, and measured particle concentrations ranged from 0 to 130 mg/L [Villholth et al., 2000]. Although some of the lowest water contents reported in these previous studies are similar to the ones in our study here, our flow rate was considerably lower. We are not aware of any study where colloid mobilization was studied and reported under such a low flow rate as the 0.05 mm/d used in our study.
 From field and laboratory studies on colloid transport and mobilization, it was reported that the first flush of irrigation or rainfall generally produced the most colloid release [El-Farhan et al., 2000; Schelde et al., 2002; Vendelboe et al., 2011] and that the amount of colloids released tends to increase with increasing water content (and correspondingly increasing flow rate) [Kaplan et al., 1993; Villholth et al., 2000; Gamerdinger and Kaplan, 2001]; however, there is not always a strong correlation between colloid mobilization and rainfall intensities or discharge rate [Biddle et al., 1995; Jacobsen et al., 1997; Ryan et al., 1998; Worrall et al., 1999]. For instance, using soil from Rocky Flats in Colorado, Ryan et al.  reported a poor correlation between particle concentrations and discharge rates. In their case, a lower discharge rate of 83 mm/h mobilized more particles than a higher discharge rate of 167 mm/h. Such results can be explained by slow particle regeneration and diffusion-limited colloid mobilization in the soil between rainfall events [Jacobsen et al., 1997; Ryan et al., 1998; Schelde et al., 2002; Majdalani et al., 2008].
3.2.3. Continuous Colloid Release and Total Amounts of Colloids Released
 Figure 6d shows a continuous particle release from the sediment core, and based on the total amount of colloids released over the experimental period, we calculated a colloid release rate of 755 mg/(m2 year). Over the total period of the experiment since colloid outflow was observed (967 days), a total of 0.392 g of particles was collected, corresponding to a cumulative particle flux of 2 g/m2. Based on this, the mobilized particles over 967 days constitute 1.1% (by mass) of the total dispersible particles.
 Compared to the low recovery rate in our study, Shang et al.  reported considerably higher colloid release rates of up to 2885 mg/(m2 hour) in packed Hanford sediments having volumetric water contents between 0.21 and 0.32 cm3/cm3. Disturbance and drying of the sediments, as well as higher water content and flow rate, likely contributed to the differences in the rates and amounts of colloids released between the study of Shang et al.  and our experiments. Disturbance has been found to promote colloid release also in other studies [Bunn et al., 2000]. As disturbance dislodges colloids physically from sediment or soil particles, the increased colloid mobilization from packed disturbed sediment is not unanticipated.
3.2.4. Effect of Experimental Boundary Conditions
 The evaporation occurring at the top of the column can cause colloids to accumulate at the air-water-solid interface line, as pointed out by Wan and Tokunaga . During that process, colloids can get pinned in thin water films, thereby becoming less mobile [Wan and Tokunaga, 1997]. Subsequent irrigation, however, will likely remobilize these colloids, therefore the effect of evaporation and irrigation will, to some degree, offset each other. The readings from the top water content and water potential sensors (at 9.5 cm depth) indicate that evaporation and intermittent irrigation did not penetrate to the 9.5 cm depth (Figures 4 and 5). We therefore believe that the artifact caused by the upper boundary condition in our system does not contribute a substantial bias in colloid mobilization.
 The failure of the bottom plate to produce suction caused the sediment core to wet up at the bottom boundary. The increased water contents due to this boundary effect could cause more colloid release than otherwise would have happened. This could cause an overestimation of colloid release from our core. However, the majority of the core was little affected by the boundary condition, as the bottom water potential and water content sensors did not reflect a pronounced boundary condition effect.
3.2.5. Model Simulations
 Model simulations of colloid release could represent the general trend of the experimental data (Figure 6d). Colloid transport parameters used for the modeling are listed in Table 3. The fitted colloid release coefficient was β = 1.80 × 10−5 d−1, with lower and upper limits of 1.45 and 2.50 × 10−5 d−1, determined by adjusting the parameter β to fully encompass the experimental data (Figure 6d and Table 3). Using the same model, Shang et al.  reported a range of colloid release coefficients from 10 to 120 d−1 for flow rates from 260 to 415 mm/d. These values are 6 to 7 orders of magnitude higher than our rate. This large difference in release rates indicates that particle release is considerably smaller in our undisturbed core under the much lower flow rate than in the disturbed, packed column of Shang et al.  at considerably higher flow rates and water contents.
Table 3. Colloid Transport Parameters Used in HYDRUS-1D
 Figure 7 shows the EDX spectra and corresponding TEM images of outflow samples at different times. Colloidal-size particles were observed in all outflow samples. EDX spectra indicate the presence of O, Na, Al, S, Si, Cl, K, Ca, Fe, and Br. While Na, Cl, Ca, and Br are likely contributed by the irrigation solution, O, Al, Si, K, and Fe indicate the existence of silicates, aluminosilicates, and Fe-oxides or hydroxides. Exact identification of minerals was not possible because not enough material was available for X-ray diffraction. Although the particle counts and particle concentrations were low during the experiment, colloidal-size particles could be consistently found in TEM images corroborating the continuous mobilization of colloids from the sediment core.
Maher et al.  and Singleton et al.  reported weathering rates for Hanford sediments based on Sr-isotope measurements made on an uncontaminated borehole core from the Hanford 200 area. Bulk weathering rates for primary minerals (quartz, plagioclase, K-feldspar, biotite) range from 10−7.1 to 10−6.1 g/(g year) for the depth interval of 12 to 19 m [Maher et al., 2003]. The reported weathering rates correspond to the mass of minerals weathered per total mass of minerals and time. Based on these rates, we calculate the mass of minerals weathered in our core as 75.4 to 754 mg/(m2 year). Our observed colloid release rate of 755 mg/(m2 year) is at the upper range of the calculated weathering rate. Although direct quantitative comparison between weathering rates and colloid release rates are challenging, the similarity of the two rates suggests that mineral weathering can be a major source for mobile colloids in our core.
3.2.7. Mechanism of Colloid Release
 In absence of physical and chemical disturbances, colloid release from stationary surfaces into the pore fluid should be controlled by thermokinetic energy of the colloidal particles. Colloids attached to a solid surface in a secondary energy minimum can be released when their thermokinetic energy exceeds the attachment energy [Hahn and O'Melia, 2004]. In our experiment, there were no chemical disturbances as our irrigation solution mimicked the pore water chemistry; however, there were some physical disturbances due to changes in water contents. These changes, although not pronounced, may have caused colloid release. Nonetheless, we observed continuous colloid release over 3 years, during which time there were prolonged periods with steady state flow. Based on this observation, we believe that in our experiment the majority of the colloids were released from the sediments by thermokinetic processes. A similar mechanism based on diffusion was proposed by Schelde et al.  to explain the prolonged release of colloidal particles from macropore wall into the fluid phase.
 This study was conducted to investigate whether particles can be mobilized and translocated under conditions typical for the deep vadose zone in semiarid regions. While we had the premise that particle mobilization is hindered under the low water contents and low steady state flow rates in the deep vadose zone in semiarid regions, we did observe continuous particle release. Our 5.3-year study showed that in situ colloid mobilization occurred at a low flow rate of 0.05 mm/d (= 18 mm/y) in an undisturbed sediment core. Although release rate and mass recovery of particles were much lower than reported from previous studies, where colloid mobilization from disturbed sediment columns was reported, we did observe a continuous flux of particles leaving the sediment core.
 We are not aware of other studies that have reported colloid mobilization and transport in undisturbed sediments under flow rates as low as 18 mm/y, so we can not compare our findings directly with other reports. However, the rates and amounts of colloid releases in our study are several orders of magnitude smaller than reported from other colloid mobilization studies using similar sediments, and are also smaller than amounts of colloid release reported from agricultural soils. The rate of particles release was in the range of rates of mineral weathering, which suggests that mobilized particles are potentially originating from mineral weathering.
 At Hanford, water contents vary considerably among different stratigraphic units. The water contents measured in our core fall within the range of field-measured water contents for corresponding depths in the Hanford vadose zone. While some stratigraphic units have lower water contents, thereby would limit particle mobilization and transport, it appears that, in general, the water contents in the Hanford vadose zone allow particle mobilization and transport. Our results show that in semiarid regions, the thick vadose zone with its low water content and flow rates does not necessarily constitute a perfect filter for particle transport. Even under low, quasi steady state flow rates, particles were mobilized. The continuous particle mobilization observed here may be a possible pathway for colloid-facilitated contaminant transport at the US DOE Hanford site.
 This material is based upon work supported by the US Department of Energy, Office of Science (BER), under Award No. DE-FG02–08ER64660. We thank Bruce Bjornstad for help with the sampling of the undisturbed core. We thank, Jeff Boyle for elecrophoretic mobility determinations, Chris Davitt and Valerie Lynch-Holm for help with TEM and EDX, and the WSU Franceschi Microscopy and Imaging Center for access to their facility. We thank the associate editor and two reviewers for their helpful comments, and we particularly acknowledge the associate editor's suggestion to compare colloid release rates with weathering rates.